JP2016539757A - Method for reconstructing a 3D image from a 2D X-ray image - Google Patents

Abstract

The present invention relates to a method for reconstructing a 3D image from a 2DX-ray image acquired with an X-ray imaging system, the method comprising: a) receiving a set of 2DX-ray images of a patient region with the X-ray imaging system; B) calculating an initial 3D image within the coordinate system of the X-ray imaging system by using at least a portion of the 2DX X-ray images comprising their respective projection geometry data; and c) at least one of the 2DX X-ray images. Projecting an initial 3D image onto the part and adjusting the respective projection geometry data of the image, the adjustment being the position of the image with the projection of the initial 3D image using an inter-image registration technique And d) compute a 3D image that is updated with a complete set of 2DX ray images with their respective adjusted projection geometry data. Includes a step that, a.

Description

  The present invention relates generally to medical X-ray imaging systems, and more particularly to a method and apparatus for improving the quality of 3D (three-dimensional) images reconstructed from a series of 2D (two-dimensional) projection images. .

  X-ray imaging systems (X-ray imaging systems) are frequently used during medical surgical procedures and interventions to provide doctors with image-based information about anatomical conditions and / or surgical instrument position and orientation. .

  These devices typically provide a two-dimensional projection image with different structures superimposed along the x-ray path.

  A typical example of such a device for use in an intraoperative setting is a so-called C-arm that is used movably or statically and consists essentially of a base frame, which has an intermediate joint ( A C-shaped arm with a joint is attached, and the intermediate joint allows the C-shaped arm to move in space along several degrees of freedom.

  One end of the C-shaped arm supports the X-ray source and the other end supports the image detector.

  Because of the limited information provided by these 2D images, 3D imaging techniques have become essential over the past decades.

  Computed tomography is an established class of static X-ray imaging systems used for 3D reconstruction in the radiology department, but these devices are generally not usable inside the operating room.

  In recent years, there has been increasing interest in tomographic reconstruction techniques, also known as cone beam reconstruction techniques using two-dimensional detectors. For example, background information on these reconstruction techniques can be found in Non-Patent Document 1 and Non-Patent Document 2.

  The above-mentioned C-arm can automatically acquire a series of 2D images and provide 3D information by subsequent 3D image reconstruction based on the cone beam reconstruction technique (Non-Patent Document 3-13). Special efforts have been made to do so.

  In recent years, so-called cone-beam computed tomography systems have been introduced to create 3D images of a portion of a patient, eg, for dental applications, simply by creating a complete rotation of the image plane and source contained within the occlusion torus. Is generated. This can be seen as a special design of the C-arm.

  To obtain high-quality 3D images with respect to spatial resolution, geometric fidelity, etc., projection geometry, i.e. sources in a general referential system for 2D images used for reconstruction and It is essential to know the exact position and orientation of the detector.

  However, because of their maneuverability, C-arms that are very well adapted to intraoperative 2D imaging tasks, but are not originally designed for 3D, have sufficient accuracy due to their open gantry design. Accordingly, it is not mechanically rigid enough to reproduce the desired projection geometry along the chosen path.

  For example, the nominal trajectory of the C-arm during automatic 2D image acquisition can be easily measured using an encoder integrated into the joint.

  However, the real trajectory differs from the nominal trajectory for different reasons, as well as authentic kinematics in robotics and its nominal model. Open gantry designs tend to mechanically distort, eg, bend, the device depending on the current position / orientation. Specifically, the C-arm tends to collide with a door or other object while moving around, causing inelastic deformation of the C-arm. Depending on the type of bearing and drive of the C-arm, due to its unique mass and the mass of the X-ray source and X-ray detector, C-arm wobble cannot be avoided, changing the nominal trajectory, This introduces a spatial resolution constraint on the reconstructed 3D image as a geometric error.

  To overcome these problems, different methods are known from the literature to calibrate the C-arm imaging geometry, i.e. to measure the projection geometry for the trajectory used to acquire the image.

  As used herein, the term projection geometry encompasses the detector position and orientation as well as the x-ray source position relative to a common reference system.

  The general technique is that the projected geometry for each image taken through the scan is such that the predicted marker projection based on its geometry and phantom model is optimally matched to the marker location identified in the image. The use of a clear calibration phantom that allows an accurate determination of the whole. Such a calibration phantom, which is rather cumbersome due to its required volume expansion allowing the determination of the entire projection geometry, can be online (ie during each diagnostic use) or offline (while not in diagnostic use) Can be used.

  In practice, a common approach to off-line calibration of an imaging system consists in acquiring an image of a special calibration phantom that includes a radiopaque marker prior to diagnostic image acquisition. The phantom remains stationary during the scan. Next, the projected image is evaluated in a pre-processing step to extract marker shadow locations and correspondences from the image. This is followed by the calibration step itself, which is usually based on an estimated error metric, often the mean square error between the detected marker shadow location and the estimated marker shadow location based on the current estimation of the projection geometry. Construct an optimal estimate of the projection geometry for.

  U.S. Pat. Nos. 5,057,028 and 5,037,086 disclose such an off-line calibration phantom that consists essentially of a radiolucent cylindrical tube, so that radiopaque markers of different sizes are accurately known. It is attached to its circumference at the specified position. During the calibration process, a series of images are taken through the trajectory selected for image acquisition, and the phantom is positioned so that the marker is visible in all images without phantom repositioning. Using known image processing methods, the marker center is calculated in the projected image and labeled, i.e. assigned to, the corresponding marker in the phantom that caused the shadow in the image. If enough marker centers are calculated and assigned, then the projection geometry for all images can be computed in a generic reference system.

  Calibration using such an off-line phantom is typically performed once prior to the first clinical use of the system and then at longer or shorter intervals, eg every 6 months. To be implemented. Due to its nature, this method fully addresses reproducibility deviations.

  On the other hand, it cannot compensate for irreproducible deviations in projection geometry, such as thermal shifts, fatigue over time, mechanical deformations due to device crashes during use or transport.

  Since deviations can only be detected during off-line calibration, there is a risk that at some time between two recurring off-line calibrations, the 3D image lacks the accuracy required for clinical use. Another drawback of the offline method is that it limits the use of the C-arm for 3D reconstruction to only one trajectory, ie a calibrated trajectory.

  Different on-line calibration methods are known from the literature in order to be able to compensate for non-reproducible errors.

  These methods use a phantom consisting of radiopaque markers placed in the volume to be acquired and reconstructed as a 3D image to perform calibration during diagnostic image scanning of the device. aim.

  One common problem of such on-line calibration that aims at full calibration, ie, based on radiopaque markers that determine the complete projection geometry for each acquired image, is that the C-arm is diagnostic. The fact is that for imaging purposes, the anatomical region of interest (ROI) is positioned to be visible in each image.

  Obviously, the phantom cannot be placed exactly at the same location, so it is usually difficult if not impossible to guarantee the visibility of the phantom in all images, so many In this case, full calibration is not possible.

  Another problem is that a highly radiopaque or highly radioactive object, such as a metal part such as an operating table, surgical instrument, implant, etc., is one of the phantom markers in one or more of the images. It can block one or more.

  U.S. Patent No. 6,057,059 aims to solve this problem by positioning an online arc-shaped phantom around the patient, and operates essentially in the same manner as the offline calibration method described above. However, while providing potentially good calibration accuracy, this phantom significantly impedes surgical access to the patient, is very cumbersome and expensive to manufacture and maintain high mechanical accuracy .

  To overcome these problems, other methods for on-line calibration have been proposed.

  U.S. Patent No. 6,057,034 is an on-line based optical camera attached to both ends of a C-shaped arm that allows the determination of the position and orientation of both the detector and source relative to an optical marker ring positioned around the operating table. A method for calibration is disclosed. The marker ring required in this method complicates patient access and must be repositioned each time the C-arm position or orientation is adjusted.

  U.S. Patent No. 6,057,051 discloses a method for on-line calibration that uses an ultrasonic or electromagnetic tracking system to capture the position and orientation of both the source and detector. This system tends to cause occlusion and magnetic field distortion problems and ergonomic constraints that prevent these systems from being used in a clinical environment.

  Patent Document 6 discloses a method for on-line calibration based on a strain gauge that measures the warpage of the C-arm. This system is complex to manage and maintain with a high degree of accuracy over time, which cannot recover the full projection geometry and only capture some deformation.

  U.S. Patent No. 6,057,034 discloses a calibration method that eliminates any additional measuring instrument by searching for and correcting trajectory deviations visible in sinograms using natural features in projected images. Is disclosed. While this method is theoretically effective, it has not been clarified how the image features to be tracked in the sinogram are identified. No method has been described for detecting such natural features with a high degree of accuracy and reproducibility, which makes the method practically unusable.

US Pat. No. 6,715,918 US Pat. No. 5,442,674 US Pat. No. 6,038,282 US Pat. No. 6,079,876 US Pat. No. 6,120,180 German patent application No. 1013934 US Pat. No. 7,494,278 US Pat. No. 6,370,224 US Pat. No. 7,672,709

Feldkamp, L C. Davis, LW. Kress. Practical cone-beam algorithm. J. Optical Society of America. A 1984, 1, pp: 612-619; Yves Trouset, Didir Saint-Felix, Anne Rougee and Christine Chardenon. Multiscale Cone-Beam X-Ray Reconstruction. SPIE Vol 1231 Mecial Imaging IV: Image Formation (1990); B. D. Smith, “Image reconstruction from cone-beam projections: necessary and sufficient conditions and reconstruction methods”. Grass, M. et al. Three-dimensional reconstruction of high contrast objects using C-arm image intensifier projection data.Computerized medical imaging and graphics the official journal of the Computerized Medical Imaging Society 23, 311-321 (1999). Stubig, Timo et al. “Comparative study of different intra operative 3-D image intensifiers in orthopedic trauma care.” The Journal of Trauma 66.3 (2009): 821-830. Kendoff, Daniel et al. “Limitations and pitfalls of 3-D fluoroscopic navigation in orthopaedic trauma surgery.” Technology and health care official journal of the European Society for Engineering and Medicine 17.2 (2009): 133-140. Print. Wndl, K et al. “Iso-C (3D0-assisted) navigated implantation of pedicle screws in thoracic lumbar vertebrae.” Der Unfallchirurg 2003: 907-913. Print. Siewerdsen, J. H. “Cone-bean CT with a flat-panel detector on a mobile C-arm: preclinical investigation in image-guided surgery of the head and neck.“ Proceedings of SPIE 5744.II (2005): 789-797. Siewerdsen, J. H. et al. “Volume CT with a flat-panel detector on a mobile, isocentric C-arm: pre-clinical investigation in guidance of minimally invasive surgery.” Medical Physics 32.1 (2005): 241-254. Nakashima, Hiroaki at al. “Comparison of the percutaneous screw placement precision of isocentric C-arm 3-dimensional fluoroscopy-navigated pedicle screw implantation and conventional fluoroscopy method with minimally invasive surgery.” Journal of spinal disorders techniques 22.7 (2009): 468- 472. Powell, Michael F, Diane DiNobile, and Arra S Reddy. “C-arm fluoroscopic cone beam CT for guidance of minimally invasive spine interventions.” Pain Physician 2010 51-59. Kotsianos, Dorothea et al. “3D imaging with an isocentric mobile C-arm comparison of image quality with spiral CT.” European Radiology 14.9 (2004): 1590-1595. Hufner, Tobias et al. “Utility of intraoperative three-dimensional imaging at the hip and knee joints with and without navigation.” The Journal of Bone and Joint Surgery 91 Suppl 1 (200): 33-42. LixuGu, Terry Peters. “3D Automatic Fiducial Marker Localization Approach for Frameless Stereotactic Neuro-surgery Navigation.” Proceedings of the International Workshop on Medical Imaging and Augmented Reality MIAR (2004), pp. 329-336. Bos, C. B. and Amir I. (1990): Design of fiducials for accurate registration using machine vision, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 12, n. 12, pp. 1196-1200. P. Besl and N. McKay.A method for Registration of 3-D Shapes.IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 14 (2): 239-256, February 1992. M.P. Deshmukh & U. Bhosle, A survey of image registration.International Journal of Image Processing (IJIP), Volume (5), Issue (3), 2011. Oliveira FP, Travares JM. Medical image registration: a review.Comput Methods Biomech Biomed Engin. 2012 Mar 22.

  Accordingly, the present invention aims to improve the quality of the reconstructed 3D image by overcoming at least one of the above disadvantages.

  The present invention provides 3D image reconstruction quality for mobile X-ray imaging systems used for intraoperative 3D imaging, particularly in orthopedic surgery, trauma, and other surgical or diagnostic fields. Provide a way to improve.

The method is
a) receiving a series of 2DX-ray images (a set of 2DX-ray images) of a patient's region with an X-ray imaging system;
b) computing an initial 3D image within the coordinate system of the X-ray imaging system by using at least a portion of the 2DX X-ray images comprising their respective projection geometry data;
c) projecting an initial 3D image onto at least a portion of the 2DX ray image and adjusting the respective projection geometry data of the image, the adjustment comprising initial 3D using an inter-image registration technique; Including image alignment with image projection;
d) computing an updated 3D image with a complete set of 2DX ray images comprising their respective adjusted projection geometry data;
including.

  By calculating the initial 3D image and using the projection of the initial 3D image for registration between the 2DX line image and the projection of the initial 3D image, it is possible to adjust the projection geometry data of the 2DX line image.

  Therefore, the adjusted projection geometry data takes into account non-reproducible positioning deviations of the source and detector of the X-ray imaging system that are not handled (not covered) by the nominal projection geometry data.

  In that case, the adjusted projection geometry data may be used to compute an updated image with greater accuracy.

  The above steps may be repeated until the required accuracy is achieved.

  The method described above provides a general solution for on-line calibration of C-arms that can be used practically and practically, thus significantly improving the accuracy and sharpness of 3D image reconstruction over known methods. Bring. As such, the present invention is a known problem in today's devices: lack of sufficient accuracy and fidelity of 3D reconstruction, or clinical phantoms currently available for improved on-line calibration. Overcoming the lack of applicability. In addition, the present invention compensates not only for the problem that the imaging device is slightly deformed relative to its nominal calibration, but also for some of the patient movements that may occur during image acquisition, eg due to breathing. It is also intended to solve the problem that must be done.

  Several markers are used to enhance the robustness and accuracy of the method, which can be used even if the nominal parameters are known with insufficient accuracy. The present invention is based on information provided by a variable amount of radiopaque markers that can be detected within a variable subset of the complete set of acquired 2D images, according to the present invention, from a design, encoder or offline calibration method. By adjusting the known nominal projection geometry data, it is possible to increase the quality or volume of the reconstructed 3D image by improving the projection geometry data for each acquired 2D image. In that specific embodiment, at least one marker is required. In this sentence, “detectable” means that each marker can be automatically detected by software in a given image. This is because it is in the field of view for the image and is not partially or totally obscured by body structures such as bone. Thus, a marker may be detectable in some images but not in other images.

  In a first aspect of the invention, 3D reconstruction uses a set of 2DX-ray images of the patient's region, and a calibration phantom that includes radiopaque markers is examined during acquisition of the set of 2DX-ray images. Placed in close proximity to the anatomical region of the patient.

  The phantom may be attached to the patient invasively, eg, via a clamp, screw or pin, or non-invasively, eg, using an adhesive patch. Radiopaque markers can vary in number (requires at least one radiopaque marker), and radiopaque markers can be of different shapes, such as a ball shape or a cylinder shape.

  In one embodiment, the radiopaque marker is ball-shaped.

  In other embodiments, the radiopaque marker is needle shaped.

  In one embodiment, the calibration phantom includes a radiopaque marker that is covered by a reflective material that is detected by an optical localizer.

  In one embodiment, the calibration phantom includes a surgical implant.

  In other embodiments, the radiopaque marker is constituted by an electromagnetic coil, which is used to configure a transmitter or receiver of an electromagnetic localization device that is implanted within the surgical navigation system.

  In other embodiments, the radiopaque marker is a combination of a ball-shaped marker, a needle-shaped marker, and / or an electromagnetic coil.

  By using a support structure built into the calibration phantom, the markers have a certain spatial relationship to each other, which is known by design or by metrology. Such a metrology measurement is preferably performed before the first use as well as at repetitive intervals. In another preferred embodiment, such a metrological measurement of the relative position of the marker is made at the very beginning of image acquisition using a 3D localization device (optical or electromagnetic) Such a 3D localization device is a component of any standard navigation system.

  A typical x-ray imaging system targeted by the present invention includes an x-ray source and image detection supported by a mobile arm with multiple degrees of freedom to position and orient the x-ray source and image detector in space. And a data processing unit.

  During scanning, a set of 2D images is acquired, each 2D image corresponding to a slightly different position of the source-detector pair.

  In addition to the anatomy under examination, X-ray shadows collected by phantom markers appear on a subset of acquired 2D images, whereas the number of 2D images containing marker shadows detected by the software is Depending on the geometry of the phantom and its position relative to the trajectory of the source-detector pair during operation.

  On the other hand, at least one 2D image does not contain any automatically detectable markers.

  After scanning, in the first step, all images containing markers that can be detected automatically by the software are selected and the positions of all detected markers in those images are precisely calculated by the data processing unit. The Such images containing automatically detected markers are identified as reference images, while the remaining images are considered non-reference images.

  The nominal projection geometry data of the X-ray imaging system is then used so that the distance between the projection of the 3D position of the marker in each reference image and the corresponding 2D position in the reference image is minimized. X-ray imaging system by optimizing the alignment between the known 3D position of at least one marker of the calibration phantom and the corresponding 2D position of said marker detected in at least two different reference images An optimal rigid transformation between the current coordinate system and the calibration phantom coordinate system is computed.

  The optimal isometric transformation is applied to the 3D position of the at least one marker of the calibration phantom to determine its respective transformed 3D position in the coordinate system of the X-ray imaging system.

  For each of the reference images, the projected geometry data adjusted so that the projection of the transformed 3D position using the adjusted projection geometry data is optimally matched with the 2D position of the corresponding marker. It is calculated from the 2D position of at least one marker detected in the image and the converted 3D phantom marker position.

  Next, a reconstructability criterion characterizing the ability to reconstruct a 3D image with sufficient quality from only the reference image is calculated.

In step b), the computation of the initial 3D image in the coordinate system of the X-ray imaging system is
-By using a reference image with their respective adjusted projection geometry data if the reconstruction criteria are satisfied, or-their respective adjustments if they do not satisfy the reconstruction criteria By using reference images with projection geometry data and non-reference images with their respective nominal projection geometry data,
Implemented

  In step c), an initial 3D image is projected onto the non-reference image, and the respective nominal projection geometry data of the non-reference image is the non-reference to the projected 3D image using an inter-image registration technique. Adjusted by image alignment.

  In step d), the computation of the updated 3D images uses a complete acquisition set of 2DX-ray images comprising their respective adjusted projection geometry data.

  According to one embodiment, one or several of the above steps are repeated until the alignment quality measurement is below a predetermined threshold or until a predetermined number of iterations is reached.

  When the selected step is repeated, the new iteration will be in the coordinate system of the X-ray imaging system after its previous alignment step has been performed to improve the number of automatically detected markers. The projection of at least one marker of the phantom, known as

  According to one embodiment, the optimal isometric transformation is the best between the 2D position of the marker detected in the reference image and the 3D phantom marker position that is backprojected according to the nominal projection geometry data. Calculated as a fit.

  According to one specific embodiment, only at least one of the calibration phantoms is detected on at least two 2D images, and the optimal isometric transformation is a translation.

  According to one embodiment, the change applied to the nominal projection geometry data for each reference 2D image consists of a change in the position of the X-ray source in a plane parallel to the image detector.

  According to one embodiment, the adjustment applied to the nominal projection geometry data of the non-reference image in step (c) is limited to translation and rotation in a plane parallel to the image detector.

  According to one embodiment, at least one radiopaque marker is automatically detected in at least two of the acquired 2DX ray images, which form an angle of at least 15 ° with respect to each other.

  According to one embodiment, the calibration phantom includes only one or two radiopaque markers.

  According to one embodiment, the calibration phantom includes a localization element that allows navigation of the surgical instrument in the reconstructed 3D image.

  Further features, advantages, and advantages of the present invention will become apparent from the detailed description that follows with reference to the accompanying drawings.

FIG. 2 schematically illustrates an exemplary implementation of X-ray imaging with multiple degrees of freedom. FIG. 1 schematically illustrates an exemplary embodiment of an X-ray imaging system comprising a phantom attached to a patient under examination. FIG. 2 schematically illustrates an exemplary embodiment of a phantom with a radiopaque marker and clamp fixation. FIG. 2 schematically illustrates an exemplary embodiment of a phantom with a radiopaque marker and pinning. FIG. 6 schematically illustrates an exemplary embodiment of a phantom with a radiopaque marker and an adhesive pad fixation. FIG. 2 schematically illustrates an exemplary embodiment of a phantom comprising a radiopaque marker and an arcuate support. FIG. 6 schematically illustrates an exemplary embodiment of a phantom with two ball-shaped radiopaque markers having different diameters. FIG. 2 schematically illustrates an exemplary embodiment of a phantom with one ball-shaped radiopaque marker.

  FIG. 1 a depicts one embodiment of a mobile X-ray imaging system 20 that includes an X-ray source 22, an X-ray detector 24, a C-shaped arm 26, and a chassis 28 with attached rollers 30. doing.

An X-ray source 22 is attached to one end of a C-shaped arm 26 and an X-ray detector 24 is
The other end of the C-shaped arm 26 is attached.

  The X-ray source 22 and X-ray detector 24 are mounted facing each other at the opposite end of the C-shaped arm 26 in a manner known in the art.

  The 2D image captured by the detector 24 may be sent to a graphic display not shown in the drawing for direct visibility.

  The 2D image may be stored in the memory of the data processing unit of the X-ray imaging system.

  The data processing unit (not shown) is specifically configured to compute a 3D image from a set of acquired 2D images, compute a transformation between different coordinate systems, etc. Typically, it includes a processor.

  The data processing unit further includes a memory for storing acquired images, imaging system parameters (specifically, nominal projection geometry data), and results of computations.

  As further illustrated in FIG. 1 a, the C-shaped arm 26 is attached to the holder 32, and the holder 32 moves along the circumference of the C-shaped arm 32 while the C-shaped arm 26 slides within the holder 32. Movement, thereby achieving orbital rotation as indicated by arrow 34.

  Holder 32 is linked to horizontal guide 36 to allow horizontal translation of C-shaped arm 26 as indicated by arrow 44 and angular rotation about angular rotation axis 38 as indicated by arrow 40.

  The horizontal guide 36 is attached to the vertical column 42 for the vertical translation (translation) movement of the C-shaped arm 26 as indicated by the arrow 46.

Table 1 summarizes the degrees of freedom that the imaging system offers using its different joints.

  It should be noted that it is not important to the present invention for the imaging system to accurately perform the above kinematics with its associated degrees of freedom, but rather many possible architectures with multiple degrees of freedom can be selected. is there.

  For example, the imaging system can include additional rotational degrees of freedom around the vertical column 42, which has no effect on the present invention. The imaging system may also include translation of the mobile device on the floor at a right angle to the C-shaped plane.

  An architecture in which the function of the C-shaped arm is performed by a multi-axis robot such as a Syngo device from Siemens, or a simple cone beam computed tomography used in dentistry such as the i-CAT system from Imaging Science International, for example. It is also possible to use a modulo (CBCT) architecture.

  Some or all of the joints include position encoders (not shown) as are known in the art. It is also possible to use several global tracking systems consisting of gyroscopes and accelerometers in combination with or instead of encoders.

  These encoder measurements may be transmitted to the data processing unit of the X-ray imaging system 20.

  In particular embodiments of the present invention, some or all of the joints may be locked (locked) by a brake, either individually or in combination.

  In certain embodiments, the adjustment of the different joints is performed using a motor not shown in the drawings.

  The motor may be controlled by a data processing unit.

  According to a preferred embodiment, a calibration phantom is used to provide a detectable marker on at least a portion of a set of 2D x-ray images.

  FIG. 1 b depicts a calibration phantom 48 that is attached to a patient 50 lying on a patient table 52.

  The mobile X-ray imaging system 20 is positioned relative to the patient table 52, the patient 50, and the phantom 48 so that the operation of the imaging system 20 during scanning is not hindered, i.e., no collision occurs.

  The calibration phantom has the following components.

  One or more radiopaque markers made from steel, titanium, tantalum, or aluminum with a known shape, such as a ball, cylinder, thin tube, or combinations thereof Constitute functional components for the calibration process. The marker may have a complex shape corresponding to a known model of a surgical instrument or implant.

  In one preferred embodiment, all markers are made of, for example, polyetheretherketone (PEE) or Plexiglas® or the like in order to build and maintain a certain known spatial relationship with the markers. Rigidly attached to a substantially radiolucent support structure, such as carbon fiber. The position of all markers is accurately known by metrology or design in a coordinate system 49 linked to the calibration phantom. Such a metrology measurement is preferably made before the first use and at repeated intervals. In another preferred embodiment, such a metrological measurement of the relative position of the markers is performed only at the start of image acquisition using a 3D localization device (optical or electromagnetic). Once done, such a 3D localization device is a component of any standard surgical navigation system.

  Calibration phantoms allow differences in marker shape and size to distinguish them from the shadows they generate in the X-ray projection image, along with known relative positioning of the markers relative to each other As designed. A number of methods have been described in the literature for automatically detecting and identifying markers on X-rays or video images.

  In certain embodiments, the calibration phantom further includes a fixture that attaches a support structure that supports the marker to the patient during the scan. The fixture creates a stable spatial relationship between the anatomy to be examined through the scanning organ and the X-ray calibration phantom. As will be described below, the fixture has various embodiments and shapes that can be selected by the practitioner depending on the area of the patient that must be imaged and based on consideration of the invasiveness of the fixation solution. May include.

  In certain embodiments, the calibration phantom further includes a localization component that is rigidly connected to the support structure, and using the localization system, the space of the phantom and along with it The position of the internal coordinate system 49 can be detected. This allows 3D reconstruction to be directly correlated to a surgical navigation system as described, for example, in US Pat.

  Localization systems are based on different technologies used alone or in combination, such as optics with passive reflective markers, optics with active radiation markers, electromagnetics, gyroscopes, accelerometers, RFID, ultrasound, etc. Well, the localization components are correspondingly different.

  In one preferred embodiment, the calibration phantom includes an electromagnetic coil that is a transmitter and / or sensor of a localization device that is part of the surgical navigation system. Thus, a surgical instrument with an electromagnetic sensor or transmitter can be directly localized with respect to a coordinate system attached to the calibration phantom. Using the method of the present invention, the reconstructed 3D image is known in the coordinate system attached to the phantom. Thus, many possible 2D or 3D visualization forms can be used to display and navigate the surgical instrument position and orientation directly on the reconstructed 3D image. Additional spherical or linear markers may be added to enhance the accuracy and ease of automated marker detection and labeling. In the preferred embodiment, the calibration phantom includes a localization element that allows navigation of the surgical instrument within the reconstructed 3D image. The same principle applies to optical localizers.

  FIG. 2 schematically illustrates an exemplary embodiment of the calibration phantom 48, where the fixture comprises a clamp 54.

  A spherical radiopaque marker 56 is attached to the support structure 58, which itself is connected to the clamp 54.

  The clamp 54 is attached to the vertebral body 60 and establishes a rigid spatial relationship between the calibration phantom and the vertebral body 60.

  The localization component 59 includes an electromagnetic coil and is connected to an electromagnetic localization system (not shown), thereby allowing the position of the phantom 48 to be determined.

  FIG. 3 schematically illustrates an exemplary embodiment of the calibration phantom 48, where the fixture is comprised of a plurality of pins 62.

  A radiopaque marker 64 is attached to the support structure 66, which further includes a feed hole 68 for attaching the phantom 48 to the patient 50 using the pin 62. Multiple feed holes are used, the pin has a sharp tip that is simply inserted into the patient's bone by applying pressure, like a fold in the wall, and the rigidity is increased by multiple pins Created. Support structure 66 may include an electromagnetic sensor or transmitter that is not presented. FIG. 4 schematically illustrates an exemplary embodiment of the calibration phantom 48, where the fixture is comprised of an adhesive patch 70.

  A radiopaque marker 72 is attached to the support structure 74. An adhesive patch 70 that is secured to the support structure 74 may be adhesively attached to an anatomical structure, eg, the skin of the patient 50.

  A number of spherical markers 72 are presented.

  In one preferred embodiment, some or all of the markers 56, 64 or 72 are covered by a reflective material to be tracked by the optical localizer of the surgical navigation system. In another preferred embodiment, the spherical marker 56, 64 or 72 is a modified reflective sphere commonly used in optical surgical navigation, for example, a metal sphere (by Northern Digital Inc. of Ontario, Canada). Inserted into the center of a standard navigation plastic sphere covered with retroreflective tape (as manufactured).

  In other preferred embodiments, some or all of the markers 56, 64 or 72 may be standard reflective plastic, non-radiopaque balls, but the posts that normally hold them are Including a metal ball that perfectly coincides with the center of the reflective ball when the ball is fastened thereon.

  In other embodiments, markers 56, 64 or 72 are provided only for calibration, and they include standard plastic spheres (usually non-radiopaque) such as 66 or 74. It forms part of a rigid body.

  In a preferred embodiment, the calibration phantom includes only three or four reflective spheres that use the optical localizer of the surgical navigation system to define unique rigid body positions and orientations, and the interior of each reflective sphere is as described above. One of the techniques is used to include radiopaque microspheres. If it is expected to obtain a C-arm calibration as described in this invention and if it is expected to obtain conventional tracking when used simultaneously in an optical surgical navigation system, then this is Presents the minimum structure for a phantom.

  However, in one preferred embodiment, only one marker is secured to the adhesive patch, which makes it very simple to use and inexpensive to manufacture.

  In another preferred embodiment, two markers are secured to the adhesive patch and placed in a line parallel to the body axis so that the markers do not overlap when multiple 2D images are acquired around the body axis. Is done.

  The use of a small number of markers that are secured to adhesive patches that are secured to the patient's skin has the advantage of being non-invasive and is therefore intended to be used without surgical intervention or surgical navigation. Can be used in purely diagnostic tests that are not done.

  In that case, the accuracy of the method depends on the relative stability between the part of the skin to which the adhesive patch is fixed and the anatomy of interest.

  An adhesive patch that is secured to the patient's back is very suitable for obtaining an accurate image of, for example, a vertebra, even though the patient breathes during image acquisition.

  This feature applies to any embodiment of the present invention and presents significant advantages.

  If radiopaque features are used that are close and relatively fixed to the area to be imaged, the accuracy of the 3D reconstruction is improved in the area, even in the presence of patient breathing or motion.

  Of course, this feature has some limitations, and the final accuracy and patient motion compensation depends on the number of reference images in which the marker is detected and can be registered using inter-image registration techniques. Depends on the characteristics of the histogram for the image being rendered.

  FIG. 5 illustrates a preferred embodiment of a calibration phantom 48 that includes a support structure 78 shaped like a frame that is bent so that its curvature substantially matches the curvature of the patient's 50 back. The principle can be extended to any anatomical part of the body to adapt to its external shape. In this case, the above-described fixture may be omitted. A radiopaque marker 80 is attached to the support structure 78. The advantage of this phantom design is that the region of interest (ROI) is surrounded by radiopaque markers while maintaining the ROI itself freely, thereby reducing the number of markers that are close to the ROI and imaged during scanning. The fact is that it allows unobstructed surgical access to the ROI by the surgeon while maximizing.

  In theory, a phantom with only one marker detectable in at least two images may already increase the quality of the resulting 3D reconstruction, but using more phantom markers Increase the accuracy of calibration, thereby increasing the quality of the reconstructed 3D volume.

  On the other hand, the possibility of overlapping markers in a 2D X-ray image is increased simultaneously.

  Their useful numbers are limited because fully or partially overlapping kerma shadows make them more difficult or impossible to identify and accurately detect their position.

  In a preferred embodiment, at least half of the acquired images contain at least five detectable markers and a small percentage of the images are non-reference images. In a typical phantom design as shown in FIG. 3, a reasonable number of markers 64 is between 5-15. They are arranged on the support structure 66 so that the coordinates along the axis 65 are different for each marker. For a typical orbit scanning operation, the phantom is positioned so that axis 65 is essentially parallel to the orbital rotation axis of imaging system 20. Even if both axes are not perfectly parallel, overlapping marker shadows can be avoided.

  During the 3D scan, a set of 2DX ray images is taken at different predetermined positions of the imaging system 20 relative to the patient 50.

  In one preferred embodiment, the operation of the imaging system 20 during scanning is pure motor-driven orbit rotation with all other degrees of freedom locked, including floor movement using rollers 30. In contrast, the predetermined position corresponds to an equidistant orbital joint position measured by a position encoder.

  However, the scope of the present invention is not limited to this scanning operation, it is important to mark that it operates similarly for multidimensional scanning operations and non-equal distance scanning positions.

  For each given position corresponding to a series of position encoder values (a set of position encoder values) for all joints of the imaging system 20, the position of the focal point of the X-ray source 22 and the position and orientation of the image detector 24 are clinical. Determined in a coordinate system 29 linked to the chassis 28 (see FIG. 1b) prior to general use and stored in a look-up table (LUT).

  Known by design, metrology or off-line calibration, resulting in nominal projection data, i.e., without taking into account changes in the projection system's default projection geometry data from the actual geometry during image acquisition, and These data can be collected in different ways, resulting in data describing the initial projection geometry of the imaging system caused by material fatigue, mechanical deformation, vibration, mechanical play, and the like.

  In one preferred embodiment, these nominal projection geometry data are determined prior to clinical use of a C-arm with an off-line calibration phantom, as is known in the state of the art.

  Using such a calibration method has the advantage of taking into account reproducible positional deviations of the X-ray source 22 and the image detector 24 for values that are only known by design.

  As will be described, the present invention is advantageously implemented using markers.

For this reason, a calibration phantom including a radiopaque marker is used. In such an embodiment, generating a 3D image includes the following steps.
a) In an initial step not intended to be handled by the present invention, the calibration phantom 48 so that the radiopaque marker is close to the ROI, ie the anatomical region whose 3D image is to be reconstructed. Attach to the patient.

  At least two 2D x-ray images must contain at least one detectable radiopaque marker, but typically more than two images contain more than one marker. Including.

  However, in the case of only two images, each containing only one detectable marker, the image is sufficient for the second image relative to the first image to improve 3D reconstruction by calibration. Must be sufficiently angled with respect to each other to provide projective out-of-plane information about correct projections. In practice, the angle between the two images must be at least 15 °.

  Next, the imaging system 20 is positioned and its joints are for scanning without any collision between the imaging system and the patient, phantom, patient table, or any other physical object present in the vicinity. It is adjusted so that the required orbital rotation around the required ROI can be performed. Note that the trajectory rotation motion can be replaced with a complex trajectory for the C-arm.

  b) Next, a set of X-ray projection (2D) images is acquired at the predetermined position during the orbital rotation of the imaging system 20. The set of images may be stored in a memory of the data processing unit.

  c) In the next step, the coordinates of the center of the object at the gray level using all the pixels belonging to the object (weighted by their gray level value) (or simply above the threshold assigned to a fixed value) Using a standard image processing technique that calculates, search for all 2D images for a marker shadow that allows the data processing unit to automatically identify the corresponding phantom marker. For example, for a ball-shaped marker, the data processing unit checks the image for the occlusion contour and evaluates the roundness and size of the identified occlusion contour. Further details can be found, for example, in US Pat. Similar techniques are used to detect the center of a line or curve corresponding to the projection of a marker consisting of a cylinder or tube (ie, the center of the line or curve). Well-known techniques are used to detect the edges of the transmitter or electromagnetic sensor coils that make up the marker.

  Calculation of the exact position of the identified marker shadow in the 2D image is performed by the data processing unit by computing its center as described, for example, in Non-Patent Document 14 and Non-Patent Document 15. Is done.

  At least one of the markers can be automatically detected, i.e. for at least one of the phantom markers, a series of pixels in the image can be assigned to it, and its position can be determined. called images).

  The remaining images, that is, images for which markers cannot be detected automatically are called non-reference images. Indeed, for several reasons, not all markers are actually identifiable in all of the acquired series of images, so all projected images are reference images. Not exclusively. Taking this phenomenon into account is the basis of the present invention.

  First, one or more markers may be completely or partially outside the detector field of view for one, more, or all projection images.

  Second, one or more of the markers may overlap each other in one or several of the projected images.

  Third, one or more markers may be more than one or more of the projected images, such as metal surgical instruments, operating tables, patient dense structures such as bones, etc. In many cases, it can overlap with other high-density objects. The use of contrast agents reinforces this difficulty.

  Fourth, one or more markers may be image regions that have been overexposed or blurred by X-rays or that have a low signal-to-noise ratio, which detects them Make it impossible.

  For each reference image, the 2D position of each identified marker in that image is calculated using the image processing techniques described above.

  In the next step, an optimal rigid transformation between the coordinate systems 29 and 49 is calculated.

  In one embodiment of the invention, this optimal transformation is achieved by adjusting the standard six transformation parameters between 29 and 49 to adjust the nominal projection geometry data and calibration phantom coordinates of this image, respectively. Using the known 3D position of the corresponding phantom marker in the system, the square of the distance for all detected markers between the back projection lines calculated from the 2D position of the detected marker in each reference image Calculated using feature-based 2D / 3D reconstruction techniques that minimize the sum. Such minimization of the sum of the squares of the distances between known 3D lines and 3D points in two separate coordinate systems is either the Iterative Closest Point algorithm published in Non-Patent Document 16 or a variation thereof. Or it can be achieved, for example, using the Levenberg-Marquardt algorithm. Any known and robust algorithm that eliminates outliers can be used to improve the reliability of the method.

  In another embodiment of the present invention, this optimal transformation is performed by adjusting the 2D position of the detected marker in each reference image and each of the six transformation parameters between 29 and 49. Calculated by minimizing the distance for all detected markers between the position of the projection of the 3D position of each marker on the reference image calculated from the nominal projection geometry data. Any known and robust algorithm that eliminates outliers can be used to improve the reliability of the method.

  In one embodiment, the calibration phantom includes only signal markers. In this case, the transformation between 29 and 49 involves only translation, and then the rotation value is assigned to the zero default value.

  In one embodiment, the calibration phantom includes only two markers. In this case, the transformation between 29 and 49 involves translation and rotation about two degrees of freedom, and rotation about one axis is not determined and is assigned a zero default value.

  The reference image can be defined as an image containing at least a given number of NBs of phantom markers that can be automatically detected, where NB> 0. The higher the NB, the more accurate the adjustment applied to the nominal projection geometry data described further, but the fewer the number of reference images. Choosing the number NB depends on the specific context of the use of the invention. In some cases, the number of radiopaque markers that are normally detected in each image can be quite high. Therefore, it is recommended to select a value of NB close to the total number of markers constituting the calibration phantom. For example, if the calibration phantom consists of 8 radiopaque balls, assign a value of 7 to NB. In more complex cases, the number of radiopaque markers normally detected in each image can be very low. Therefore, it is recommended to select a value of NB that is 1 or 2.

  In a preferred embodiment, the number of markers NB defining the reference image relative to the non-reference image can be assigned a low value such as 1 for the first iteration of the generic method of the invention, and then The method can be repeated, but the number NB is increased to a higher value. This process can be repeated and repeated several times.

  d) In the next step, nominal projection geometry data for each reference image so that when projected onto the image, its associated 2D marker shadow position optimally matches the 2D position of the transformed 3D phantom marker position. Adjust.

  In one embodiment, the adjustment applied to the nominal projection geometry data for each reference image is limited to two coordinates of the x-ray source in a plane parallel to the image detector. This involves exploring two independent parameters for each 2D image. In other embodiments, adjustment of the nominal projection geometry data includes the x-ray source's three coordinates, and the full position and orientation of the image detector as presented by the matrix or six independent rotation / translation parameters, and Includes a scaling factor in the image that determines the pixel to millimeter ratio (one or two ratios depending on a priori information about the image detector (a priori information)). This involves exploring 10 or 11 independent parameters for each 2D image.

  In other embodiments, between 2 and 11, the intermediate number of parameters is adjusted for each image.

  In this way, the projection geometry is inventively corrected to compensate for the non-reproducible positional deviations of the x-ray source 22 and detector 24 that are not handled by the nominal projection geometry data.

  In the present invention we recommend the use of a fully digital flat panel image detector, which has the advantage over conventional image intensifiers of avoiding any geometric image distortion. However, if a conventional image intensifier is used, it may be necessary to consider a distortion compensation model in the nominal projection geometry data, which is usually an image expressed in pixel image coordinates and millimeters. It is a polynomial function between the coordinates in the plane. In the latter case, adjustment of the nominal projection geometry data geometry must include adjustment of the coefficients of the corresponding polynomial function, but this requires the use of many markers. Therefore, in practice, the adjustment of the polynomial function is ignored, and only the parameter corresponding to the isometric conversion as described above is adjusted. In that case, however, it is preferred to calculate a polynomial function with different coefficients in the C-arm orientation and store it within the nominal parameters. This is because the strain is known to vary considerably in the C-arm orientation (this is due to the effect of the magnetic field and mechanical deformation of the C-arm). This is done by calibrating the polynomial function for various orientations of the C-arm using a rigid planar grid fixed to the C-arm detector, taking into account a magnetometer that optimally detects the orientation of the magnetic field. Easily achieved during offline calibration. In that case, optimizing the 11 or 12 parameters of the model for each online image using the method of the present invention partially compensates for some amount of distortion variation relative to the nominal parameters. Another solution is to permanently fix a planar grid with a large number of markers on the C-arm and calculate those markers separately from the phantom markers in order to calculate the polynomial function for each online image. It is.

  e) The first 3D image can now be computed in the coordinate system 29 using standard tomographic reconstruction techniques known in the art, such as Filtered Back Projection or Algebraic Reconstruction Techniques.

  A reconstructability criterion is used to determine which projection image of the acquired series of images is used for this first reconstruction of the 3D image.

  The reconstruction criterion is determined empirically and depends on the number of reference images and the angular distribution.

The reconstructability criteria can be defined as a test that includes:
-Are there enough reference images?
-Is the angular coverage of the reference image sufficient?
-Is the angular distribution of the reference image sufficiently uniform?

  If the criteria are met (ie, the response to each of the above is “yes”), only the reference images with their corresponding adjusted projection geometry data will be used for the initial reconstruction of the 3D images. Used.

  On the other hand, if the criteria are not met (ie, the response to at least one of the above is “no”), reconstruction based only on the reference image will not give good results. For example, it is known that if the reference images are all obtained within an angular range of 90 degrees or less, the reconstruction process will result in poor quality 3D images. In a preferred embodiment, reconstruction yields not only the reference images with their corresponding adjusted projection geometry data, but also their corresponding nominal projection geometry data to yield a sufficient quality initial 3D image. This is implemented by also using a reference image comprising

  In another preferred embodiment, the projection geometry data of the non-reference image is adjusted and calculated by interpolation of the projection geometry data of the reference image surrounding the non-reference image. A 3D reconstruction is then performed using all the adjusted projection geometry data.

  In another preferred embodiment, the reconstruction is accomplished using all images using their nominal projection geometry data.

  Depending on the accuracy desired, the reconstructability criteria, and the implementation chosen, it is possible to stop the method at the end of that step. However, the method of the present invention proposes to take additional steps to increase the accuracy and quality of the reconstructed 3D image.

  f) In the next step, whether or not the 3D image is reconstructed with only the reference image, the 3D image is projected onto each non-reference image using their respective nominal projection geometry data. .

  Subsequently, 2D-2D gray levels / intensities using well-known image similarity measures, such as mutual information, joint entropy, or cross-correlation ratio, to adjust the nominal projection geometry data of the non-reference image Is applied to each non-reference image and its respective projected 3D image. For example, one of many 2D-2D alignment methods described in Patent Document 16 or Non-Patent Document 17 can be applied.

  In one embodiment of the present invention, the adjustments to the projection geometry data associated with each non-reference image are limited to image detector rotation and in-plane transformation, which includes three for each 2D-2D image alignment. Corresponds to finding independent parameters.

  In other embodiments, only the translation parameter is adjusted, which reduces the risk of incorrect adjustment.

  In other embodiments, a 2D-2D transform comprising more than three independent parameters, eg a warping transform or an affine transform that independently displaces the four corners of the image is defined. It is done.

  g) The updated more accurate 3D image is then computed using the complete series of 2D images (a set of 2D images) with the corresponding adjusted projection geometry data.

  In one embodiment of the invention, one or several steps of the above-described method may be performed until the alignment quality measurement is below a predetermined threshold (eg, the quality measurement is a 3D image contrast). Or may be sharpness) or repeated until a predetermined number of iterations (eg, one iteration or three iterations) is reached.

  In one embodiment, the adjusted projection geometry data and / or the reconstructed 3D position of the marker is used to search and detect radiopaque markers using an estimate of the theoretical position of the marker in the image. Can be repeated to start the next search. It has the advantage of increasing the number of markers detected for the next iteration of the generic method, and is therefore likely to improve the accuracy of the reconstructed image. With this mechanism, the minimum number of markers NB that define a reference image for a non-reference image can be progressively increased with each iteration.

  In other embodiments, only steps (f) and (g) are repeated to improve the quality of the final reconstructed 3D image.

  The reconstructed 3D image can then be sent to a graphic display for visualization.

  The reconstructed 3D image can also be stored in the memory of the data processing unit for further use.

  If the calibration phantom includes a localization device that is detected by a navigation system (eg, an electromagnetic sensor or transmitter or an optical tracker), the resulting 3D image is transferred directly and the 3D image Used for instrument navigation inside.

  In any of the embodiments of the present invention, the calculated adjusted projection geometry data obtained at the end of the 3D reconstruction is stored to update the previous nominal projection geometry data for further use. obtain. Further use may occur on the same day for the same patient, or may occur several days later for another patient. In that case, it is possible to track the evolution of the projection geometry data, which provides a useful indication of the maintenance of the device. For example, a sudden and substantial change in projection geometry data probably corresponds to an impact or mechanical change that requires further inspection of the device.

  In order to illustrate the method of the preferred embodiment, the method has been described for a phantom consisting of only one marker. In this example, the markers are 4 mm diameter stainless steel spherical balls (see FIG. 7). The marker 72 is simply secured to the skin of the patient 50 in the area to be imaged by using adhesive tape as for the electrocardiogram skin electrode. In this example, it is assumed that the skin does not move significantly relative to the anatomy of interest. A series of 190 images are acquired for each degree of C-arm trajectory rotation. A flat panel image detector is used. Each image is analyzed by a computer to search for a metal ball. However, due to various phenomena such as overexposure in certain areas, it is likely that the ball will not be automatically automatically detected on all images. In this example, it is assumed that the ball is not detected in the image (non-reference image) relating to the 30 degree angle segment, but is automatically detected in all other images (reference image). It is assumed that the reconstructability standard is satisfied. If the nominal projection geometry of the reference image is used, the center of the spherical ball projection is back projected in 3D. The intersection of the backprojection lines defines a 3D point that is defined as the center of the 3D spherical ball. Optimal isometric transformation between the coordinate system of the X-ray imaging system and the coordinate system of the calibration phantom translates to align the calculated center of the 3D spherical ball with the origin of the X-ray imaging system coordinates and units It is easily determined by the rotation component equal to the identity matrix. The 3D spherical ball is then projected onto each image, and translation is applied to each image to match the projection of the 3D spherical ball with the true position of the ball center on the image. Thus, the nominal projection geometry data of the reference image is adjusted by translation in the image plane. A first 3D image reconstruction is then performed using only the reference images with their adjusted projection geometry data. The resulting 3D image is then projected onto all non-reference images, and 2D-2D image registration is performed between the projected image of the 3D image and the actual image for only those non-reference images. Done. It results in translation and rotation of each image, which corresponds to three adjusted parameters of the projection geometry data. In order to accelerate the process, registration is done only by considering the central area of each real image. If the registration results in an uncertain result given by the low registration score given by the conventional criteria of the inter-image registration technique, the registration is ignored and the nominal projection geometry data is taken into account. The result of this step is a new set of adjusted projection geometry data for all images. A second 3D reconstruction is then performed using the adjusted projection geometry data applied to all images. This second projection geometry data has better accuracy and quality than the reconstruction performed using the nominal projection geometry data.

  In another embodiment, the calibration phantom consists of two spherical markers (see FIG. 6). The two balls 72 are arranged in a direction parallel to the body of the patient 50, are attached using an adhesive tape 70, and image acquisition is performed around the body. It has the advantage that the projections of both balls do not overlap each other, which produces more accuracy than using only one ball. The same method applies as before, the only difference is that (a) the search for two balls with a constant relative geometric configuration is more reliable than just one ball, and (b) The adjusted projection geometry data obtained in the first step corresponds to not only translation but also image rotation and translation. It is also possible to obtain a more accurate alignment of the ball by using only at least a square technique while maintaining only translation. As illustrated in FIG. 6, it is advantageous for the two balls to have different diameters.

  In the third embodiment, the calibration phantom consists of four balls and six pins. Six pins are used to rigidly attach the calibration phantom to the vertebrae, using percutaneous pin fixation of three pins on each side of the vertebra. The four balls are covered with a reflective material, which constitutes a rigid body that is detected by the optical localizer. This rigid body constitutes a patient reference system. The ball and pin coordinates are all known in the same unique calibration phantom coordinate system as the patient reference system. A series of 190 images are acquired for each degree of C-arm trajectory rotation. Each image is analyzed by a computer to search for metal balls and pins. In some images, at least three balls are detected and automatically labeled. They constitute a reference image. Other images constitute non-reference images. For all reference images, the detected ball and pin segments are backprojected using nominal projection geometry data. For each image, the backprojection is aligned with the 3D coordinates of the ball and pin represented in the calibration phantom coordinate system. It generates a transformation matrix between the calibration phantom coordinate system and the C-arm coordinate system defining the nominal projection geometry data. For each image, the 3D ball and pin are projected onto the image and the projection geometry data is the square of the distance between the theoretical projection of the ball and pin and their actual extraction on the image. They are adjusted to match using a least squares criterion that minimizes the sum. In this embodiment, both the x-ray source and the image origin are translated equally and, in addition, the rotation of the image plane is adjusted. Next, the remainder of the procedure is equal to the previous example. However, in that case, it is possible to navigate a surgical instrument with a reflective sphere and visualize its position in real time on the resulting 3D image. This is because the image is directly reconstructed in the coordinate system of the calibration phantom, which itself is directly the coordinate system of the patient reference system.

  In a fourth embodiment, image acquisition is performed immediately after an implant, such as a pedicle screw, is inserted into the patient's body to confirm their location. There are no additional markers attached to the patient. The calibration phantom is made directly with an implant that builds a marker having a partially known geometric configuration. Metal implants look particularly good on the image. For example, after two vertebrae have been operated on, four pedicle screws are observed on the image. The individual geometric configuration of the implant is known and stored in a computer. However, the relative geometry between the implants remains unknown. In the first step, all images that can detect and separately identify the implants are selected, and the three-dimensional position and orientation of each implant is determined using their 3D / 2D rigid registration algorithm. Reconstructed using projections detected on the image. Calibration phantoms are now completely and generally known. Then the rest of the method follows. The advantage of this method in this embodiment is to obtain a clear and precise 3D image of the bone with the implant with a high degree of precision and sharpness, which will leave the implants in their current position. Or it makes it easy to decide whether to re-operate on some of them.

  According to another embodiment of the present invention, a series of 190 images are acquired for each degree of orbital rotation of the C-arm, but no markers are detected on the 2D image. This may be due to the large volume-like characteristics that the patient creates images of poor quality. This may be true if the calibration phantom is forgotten by the user. This may be true because even though the calibration phantom is non-invasive and compact, the user simply does not want to take the time and effort to position the calibration phantom. In that embodiment, the first part of the method described above cannot be applied. This is because there are no markers or no detectable markers. In that case, it is possible to apply only the second part of the method (where all images are non-reference images). First, a 3D image is reconstructed using nominal projection geometry data associated with all acquired 2D images. Second, the reconstructed 3D image is projected onto each 2D image, still using the nominal projection geometry data. It constitutes a virtual gray level image. This projection step may benefit from a graphics processing unit (GPU), for example, using programming in the Cuda language to accelerate the process. For each 2D image, a 2D-2D image registration method is applied between the actual 2D image and the projection of the 3D image. For example, using the technique described in Non-Patent Document 17, for example, one 2D translation and one rotation (one rotation) so that the mutual information of both images is maximized within a given region of interest. An optimal isometric transformation consisting of three parameters) is searched. In order to define the region of interest, a possible solution is to search the first reconstructed 3D image for 3D pixels with high contrast after filtering the noise to identify such high contrast mass centers. And projecting the center of mass onto a 2D image, with a bounding box around the resulting projection point having an a priori size, such as half the total image size in each dimension. box) (3D image volume is divided by 8). In order to accelerate the alignment process, conventional multi-level alignment techniques may be used, for example, using wavelets or using only pyramids. For each image, three adjusted parameters are obtained, and a second 3D reconstruction is performed using all the images with their adjusted projection geometry data. Depending on the accuracy required by the application, the method can be repeated if necessary. In that case, the second 3D reconstructed image is projected on all 2D images, and 2D-2D registration is again performed between the projection of the second 3D reconstructed image and the real image. This results in a new set of adjusted projection geometry data. A third reconstruction is performed using the corresponding adjusted projection geometry data. Depending on the required regime and time constraints, the method can be repeated as many times as necessary. A fixed number of iterations can be preset, or contrast criteria can be calculated on the reconstructed 3D image. And when the criterion goes above a given threshold, the process is stopped, which has the advantage of always generating a 3D image with good sharpness.

  As mentioned above, the present invention may be practiced with or without the use of markers. Using markers enhances the stability and accuracy of the method and is required anyway if 3D images are used for navigation. Of course, it is easier for the user not to use the marker, but it means that the initial nominal projection geometry data is streamlined to the real projection geometry data so that the convergence of the method is guaranteed. Demand close. Eventually, for any part of the body, with or without a link to navigation, during diagnostic imaging or during surgery, for various X-ray system architectures, in various applications, methods Can be used. Instead of standard computed tomography, the method can also be used for non-medical applications such as non-destructive confirmation of objects using an apparatus that includes an X-ray flat panel detector. The method proposed in the present invention provides a number of parameters and options that are adjusted according to each situation.

Claims (15)

A method for reconstructing a 3D image from a 2DX ray image acquired by an X-ray imaging system,
a) receiving a set of 2DX-ray images of a patient's region with said X-ray imaging system, the calibration comprising at least one radiopaque marker having a known 3D position in the coordinate system of the calibration phantom The phantom is the acquired set,
-At least two 2DX-ray images, each comprising at least one detectable radiopaque marker of the calibration phantom;
-At least one 2DX-ray image, wherein the radiopaque marker of the calibration phantom is not automatically detectable;
To include
Being placed on the patient during the acquisition of the set of 2DX-ray images;
a1) selecting from the set of 2DX-ray images a reference image in which at least one radiopaque marker of the calibration phantom is automatically detected, and for each detected marker in each of the reference images Determining the 2D position, wherein all remaining images are classified as non-reference images;
a2) The nominal projection geometry of the X-ray imaging system so that the distance between the projection of the 3D position of the marker in each reference image and the corresponding 2D position in the reference image is minimized Data is used to optimize alignment between the known 3D position of the at least one marker of the calibration phantom and the corresponding 2D position of the marker detected in at least two different reference images. Calculating an optimal isometric transformation between the coordinate system of the X-ray imaging system and the coordinate system of the calibration phantom;
a3) applying the optimal isometric transformation to the 3D position of the at least one marker of the calibration phantom to determine its respective transformed 3D position in the coordinate system of the X-ray imaging system; Steps,
a4) for each of the reference images such that the projection of the transformed 3D position using the adjusted projection geometry data optimally matches the 2D position of the corresponding marker; Computing adjusted projection geometry data from the 2D position of the at least one marker detected in the image and the transformed 3D phantom marker position;
a5) calculating a reconstructability criterion characterized by the ability to reconstruct a 3D image with sufficient quality from only the reference image;
b) calculating an initial 3D image within the coordinate system of the X-ray imaging system by using at least a portion of the 2DX X-ray image comprising their respective geometric data, the calculation comprising:
-If the reconstructability criteria calculated in step a5) are satisfied, by using the reference image with their respective adjusted projection geometry data, or-calculated in step a5) By using the reference image comprising their respective adjusted projection geometry data and the non-reference image comprising their respective nominal projection geometry data.
The steps performed;
c) projecting the initial 3D image onto the non-reference image and aligning the non-reference image with the projection of the initial 3D image using an inter-image registration technique; Adjusting the respective projection geometry data;
d) computing an updated 3D image using said complete set of 2DX ray images comprising their respective adjusted projection geometry data;
Method.   The method according to claim 1, wherein one or several of the steps a1) to d) are repeated until the alignment quality measurement falls below a predetermined threshold or until a predetermined number of iterations is reached.   A new iteration of the selection step a1) is performed after the alignment step a2) is performed in order to improve the number of markers that are automatically detected, so that its position is within the coordinate system of the X-ray imaging system. The method of claim 1, wherein the projection of the at least one marker of the phantom is known.   The method according to claim 1, wherein the at least one radiopaque marker is ball-shaped or needle-shaped.   5. A method according to any one of the preceding claims, wherein the calibration phantom comprises a radiopaque marker covered with a reflective material detected by an optical localizer.   The method according to claim 1, wherein the calibration phantom comprises a surgical implant.   The at least one radiopaque marker includes an electromagnetic coil that can be used as a transmitter or receiver of an electromagnetic localization device that is implanted within a surgical navigation system. The method described.   The method according to claim 1, wherein the calibration phantom includes a combination of a ball-shaped radiopaque marker, a needle-shaped radiopaque marker, and an electromagnetic coil.   The step a2) is achieved by computing the transformation that results in the best fit between the 2D position of the marker detected in the reference image backprojected according to the nominal projection geometry data and the 3D phantom marker position. 9. The method according to any one of claims 1 to 8, wherein:   10. A method according to any one of the preceding claims, wherein only one marker of the calibration phantom is detected on at least two 2D images and the transformation in step a2) is a translation. .   In step a4), the change applied to the nominal projection geometry data for the computation of the projection geometry data for each reference 2D image is the X-ray in a plane parallel to the image detector 11. A method according to any one of the preceding claims, comprising the change in the position of a source.   12. In step c1), the adjustment applied to the nominal projection geometry data of the non-reference image is limited to translation and rotation in a plane parallel to the image detector. The method of any one of them.   13. In said step a), at least one radiopaque marker is automatically detected in at least two of the acquired 2DX-ray images that result in an angle of at least 15 ° to each other. The method of any one of these.   14. A method according to any one of the preceding claims, wherein the calibration phantom includes only one or two radiopaque markers.   15. A method according to any one of the preceding claims, wherein the calibration phantom includes a localization element that enables the navigation of a surgical instrument in the reconstructed 3D image.

Images